International Journal of Computer Applications (0975 – 8887)
Volume 73– No.4, July 2013
Algorithm for a PSO Tuned Fuzzy Controller of a DC
Motor
Debasmit Das
Department of Electrical Engg.
Indian Institute of Technology,
Roorkee, India
ABSTRACT
Determination of the parameters of the membership functions
of a fuzzy logic control process is the crucial factor for
providing optimum performance of the system. These
parameters are regarded as variables and are tuned through
Particle Swarm Optimization (PSO). The shape of the
membership functions vary according to the variables.
Consequently, the fuzzy control output changes and so does
the performance. The results give an insight to the efficiency
of PSO in producing optimum membership functions in real
time. This controller can be applied to various control systems
like AGC(Automatic Generation Control), DC motors etc..
Demonstration for the latter is shown in this paper.
General Terms
Swarm Optimization Algorithms, Fuzzy Controller
Optimization
Keywords
MF, MATLAB, SIMULINK, PSO, Fuzzy.
1. INTRODUCTION
For the tuning of the parameters of the membership
functions of a fuzzy controller a novel PSO algorithm has
been developed. The algorithm for the fuzzy controller has
been encoded in MATLAB but a block diagram strategy is
enabled to explain the algorithm. A SIMULINK model has
been used.The Plant used is an armature controlled DC Motor.
Conventional controllers like PI and PID controllers fail in
case of non linearities and may generate steady state error[1].
In such a case a fuzzy controller is used which is basically a
non-linear element whose parameters are tuned using Particle
Swarm Optimization Technique (PSO) subject to the
condition that steady state error is to be minimized. The
quantity to be controlled is the speed of the DC Motor.
Therefore error in speed is to be minimized. PSO technique is
a very uncertain algorithm that may or may not converge to
the optimized values. Nevertheless we got optimistic
simulation results. As such it could overcome the limitations
of conventional controllers[1].
2. FUZZY CONTROLLER AND PSO
A Fuzzy Controller is characterised by inputs and outputs.
The inputs and outputs are correlated by Membership
Functions (MF). In this case there are two inputs 1) Error of
speed. 2) Change in Error of speed.
Arka Ghosh
Department of Electrical Engg.
Indian Institute of Technology,
Roorkee, India
2.1 Membership Functions
Each of the inputs have 3 membership functions – for
Negative, Zero and Positive. The MFs are chosen to be
triangular in shape. There is only one output i.e. Voltage
which is fed to the armature of the DC Motor . The Voltage
also has 3 MFs – Negative, Zero, positive but it is not required
to be optimised. Since the input to the fuzzy controller deals
with the quantity to be controlled only the Input MF is
sufficient to be optimised. Mamdani (Fuzzy inference System)
FIS has been used. For each triangular membership function
there are 3 parameters a,b,c. Since there are 3 MFs for a
particular input there are a total of 9 parameters for a
particular input. Since there are 2 inputs so there are a total of
18 parameters for each input. The membership functions of
one input is shown.
Fig 1: Membership Functions and their Parameters
As can be seen in Fig1 a1,b1,c1 corresponds to membership
function Negative . a2,b2,c2 corresponds to membership
function Zero. a3, b3, c3 corresponds to membership function
Positive. These parameters have to be tuned such that the
steady state error is zero. Again in this case a1 ,b1 and b3,c3
will be fixed since they deal with extreme or maximum errors
or change in errors and b2 will be fixed and centred at zero.
Conditions like a1<b1<a2 and c2<b3<c3 also need to be
satisfied. a2<c1 ; c1<b2; b2<a3; a3<c2. These conditions are
dealt with by the PSO algorithm to be explicated later. So it
can be seen that there are 8 parameters to be tuned. The PSO
algorithm is required to be discussed. The PSO is an
optimization technique that optimizes a problem by
sequentially trying to improve the candidate solution. It does
not guarantee an optimal solution[2]. There is a need to
produce zero steady state error in speed by we optimising the
parameters of the fuzzy controller. In a way fixing the
objective function as a function of speed is not possible as it is
not the output of the fuzzy controller. The output of the fuzzy
controller is voltage. But the voltage more precisely has a
dependence on the speed we want. So the objective function
should include voltage rather than speed . In other words the
37

objective function for PSO should be a function of output of
the fuzzy controller. The details of how the objective function
is found out is given in the next section. Now discussion will
be on the PSO algorithm[3]. The PSO search space is
multidimensional. The dimension is determined by the
number of variables whose value we are to find. In this case
the dimension is 8. A point in the space is referred to as a
particle. Each particle moves in the search space with a
velocity dependent on its own flying experience and it’s
companions’ flying experience. Each particles keeps a note of
its best solution (best evaluating value of the objective
function) which is termed as personal best (pbest). The global
best (gbest) is the best solution in the whole group. In each
iteration the velocity and the position is updated according to
the following rule[4].
2.2 PSO Variables
For example let the jth particle be represented by x j=
(xj1,xj2,xj3,...xjg) in the g dimensional space. The best previous
position of the jth particle is pbest j=(pbestj1,pbestj2,...pbestjg).
The best particle corresponds to gbestg. The velocity of the
particle j is represented as vj=(vj1,vj2,.....vjg). The new velocity
and new position for a particular iteration k+1 is as :
vjg(k+1)=w.vjg(k)+c1*rand()*(pbestjg–xjg(k))
+c2*rand()*(gbestg – xjg(k))
xjg(k+1) = xjg(k)+vjg(k+1)
j=1,2.......n
g=1,2,.......m
where
n number of particles in a group
m number of members in a particle
Fig 2: The Simulink Model
International Journal of Computer Applications (0975 – 8887)
Volume 73– No.4, July 2013
k iteration number
v velocity
x particle co-ordiante
c1,c2 acceleration constant normally set to 2
w inertia weight factor
rand() random number between 0 and 1
In this way the particle moves about in space and for each
iteration it checks for the optimization of the objective
function. The next section will be about the working model
and the implementation of the algorithm.
3. SIMULINK MODEL AND THE
ALGORITHM
The SIMULINK model[5][6] used is given in Fig 2. As can be
seen the error and rate of change of error of speed has been
normalized and is fed into the fuzzy controller as Input [7].
The Minor Feedback loop in the model is that of a Armature
Controlled DC Motor. Hence, the input of the Minor
Feedback loop is the Armature voltage which is also the
output of the Fuzzy controller. The Minor Feedback path is
that of the Back EMF of the DC Motor. The Output MF
parameters need not be optimized. The objective function
F(i)=0.5 – Output of the Fuzzy Controller. The output of the
Fuzzy controller has been evaluated using the evalfis() of
MATLAB[8]. The value 0.5 has been calculated as per the
set point speed of 23 SI Units. Using the equation E=K*N
putting N=23 SI we get E as 0.5 and since the normalisation
constant (in the SIMULINK Block diagram) is unity. The
constant load torque has been assumed for the motor and
shown by the input to the second summer in the minor loop.
38
 International Journal of Computer Applications (0975 – 8887)
Volume 73– No.4, July 2013

Description of the algorithm for tuning the fuzzy 4. RESULTS

controller[9]: Results are obtained for two cases firstly by keeping the set
point constant and then varying the set point.
1) The no. of particles (n) is to be 30. The number of
iterations (i) is 20. 4.1 Fixing Set Point
2) Since 8 variables need to be optimised therefore the The set point of the speed is set to 23 SI units and the PSO
paricle is 8 dimensional. code is executed two times. Therefore two transient responses
3) For each particle (1 to 30) the following is done: are obtained with negligible steady state error. As expected
a) The particle’s position i.e. the parameters of oscillations occur because the presence of inductance in
MFs is initialized using the rand(). For e.g. armature circuit has been assumed. So the following tables
rand(1) gives a random number between 0 and and graphs are plotted.
1. -1+rand(1)*2 gives random number between
-1 and +1. So the lower and upper boundaries Table 1. Parameters for setpoint 23 SI Units for first PSO
of the search space are also set in this way. execution
Thus the constraints on the MF parameters can
be set
Error in speed
b) The particle’s best known position is initialized
to initial position.pi < - xi -1.8 -1 -0.0596
c) If F(pi)<F(g) the swarm’s global best position
is updated g < - pi -0.0942 0 0.1270
d) The initial velocity is set to 0. 0.0124 1 1.8
4) Until i>20 or minimum error tolerance has been
achieved the following is repeated : Error in rate of speed
a) For each particle (1 to 30) :
(i) For each dimension (1 to 8) : -1.8 -1 -0.0673
 Velocity of each particle is -0.0941 0 0.1982
calculated and updated
(ii) The particle’s position is also 0.0126 1 1.8
updated
b) If F(xi)< F(pi)
(i) Update the particle’s best known
position pi < - xi
(ii) If F(pi)<F(g) the swarm’s best
known position is updated: g < - pi
5) The M-File where the algorithm is encoded is
executed
6) Hence, the optimum values of the 8 parameters are
found and fed into a matrix. The other fixed values
are clubbed with it to form a 6 X 3 Matrix.
7) The Matrix is fed into the FIS Structure using the
function setfis().
8) The FIS Structure is uploaded into the SIMULINK
Fuzzy controller block and then the Model is
Simulated.
pi positon vector of personal best
g position vector of global best
Fig 3: Speed Response for parameters of Table 1
xi postion vector of a particle
i iteration index
Table 2. Parameters for Set Point 23 SI Units for
For the Set points of 25 and 30 SI the objective function is second PSO execution
calculated using the equation described. For Set Point 25
E=0.543 and objective function F(i F(i)=0.543 – Output of the Error in speed
Fuzzy Controller. For Set Point 30 E=0.543 and objective -1.8 -1 -0.0673
F(i)=0.652 – Output of the Fuzzy Controller. So in the
algorithm the objective function is changed for each set point -0.0942 0 0.1274
in the way mentioned above. Accordingly results are obtained
as shown in the next section. 0.0945 1 1.8
Error in rate of speed
-1.8 -1 -0.0274
-0.0339 0 0.0367
0.0345 1 1.8

39
 International Journal of Computer Applications (0975 – 8887)
Volume 73– No.4, July 2013

Table 4. Parameters for Set point 30 SI Units

Error in speed
-1.8 -1 -0.0567
-0.0630 0 0.1296
0.1143 1 1.8
Error in rate of speed
-1.8 -1 -0.0789
-0.0959 0 0.1279
0.0568 1 1.8

Fig 4: Speed Response for parameters of Table 2

4.2 Varying Set Point

Now the PSO algorithm is run for different set point. Only
single execution of the algorithm is used for the two set points
of 25 and 30 SI Units. The results obtained are given below.

Table 3. Parameters for Set point 25 SI Units

Error in speed
-1.8 -1 -0.0396
-0.218 0 0.1345
Fig 6: Speed Response for the parameters of Table 4
0.0745 1 1.8
Error in rate of speed
5. CONCLUSION
-1.8 -1 -0.0890 The PSO tuned Fuzzy controller was simulated on the plant.
-0.9800 0 0.4560 The plant was chosen to be an armature controlled DC Motor
with speed being the quantity to be controlled. The algorithm
0.0345 1 1.8 for tuning the paramaters of the membership functions of the
Fuzzy controller has been developed. The optimized
parameters so obtained is fed into the SIMULINK Block of
the fuzzy controller and results obtained for various Cases.
Steady error has been made zero. Response is oscillatory
because of presence of inductance in armature circuit. Fig. 3
shows the best response with very little overshoot.

6. ACKNOWLEDGEMENTS
We would like to thank Dr. Rajendra Prasad, Department. Of
Electrical Engineering, Indian Institute of Technology,
Roorkee, for constantly supporting and advising on our
project. Moreover we would like to thank Institute Computer
Cente, Indian Institute of Technology, Roorkee for providing
us the software MATLAB/SIMULINK without which the
project would not have been possible.

.
Fig 5: Speed Response for the parmeters of Table 3

40

7. REFERENCES
[1] S.K.Sinha, R.N. Patel, and R. Prasad “Application of GA
and PSO Tuned Fuzzy Controller for AGC of Three Area
Thermal-Thermal-Hydro Power System” International
Journal of Computer Theory and Engineering, Vol. 2,
No. 2 April, 2010
[2] Qinghai Bai “ Analysis of Particle Swarm Optimization”,
Computer and Information Science, February 2010
[3] Saeed Vaneshani and Hooshang Jazayeri-Rad,
“Optimized Fuzzy Control by Particle Swarm
Optimization Technique for Control of CSTR,” World
Academy of Science, Engineering and Technology 59
2011
[4] Zwe-Lee Gaing, “A Particle Swarm
OptimizationApproach for Optimum Design of PID
Controller” IEEE transactions on Energy Conversion,
Vol 19, No. 2, June 2004
IJCATM : www.ijcaonline.org
International Journal of Computer Applications (0975 – 8887)
Volume 73– No.4, July 2013
[5] Dipraj, and Dr. A.K. Pandey, “Speed control of a D.C.
Servo Motor By Fuzzy Controller ,” International Journal
of Scientific & Technology Research Volume 1, Issue 8,
September 2012
[6] Mehmet Cunkas, Omer Aydogdu “ Realization of Fuzzy
Logic Controlled Brushless DC Motor Drives using
MATLAB/SIMULINK”, Vol.15, No.2, 2010
[7] J.V. de Oliviera “Semantic Constraints for membership
function optimization”, IEEE transactions on Systems,
Man and Cybernetics, January 1999
[8] R.N. Patel, S.K.Sinha, and R. Prasad, “Design of a
Robust Controller for AGC with Combined Intelligence
Techniques,” World Academy of Science, Engineering
and Technology 21 2008
[9] The Wikipedia website. [Online]. Available:
http://www.wikipedia.org/
41